A modest proposal for granted credits...
18 Mar 2005 4:58:46 UTC
Topic 188591
(moderation:
... waiting to be shot down.
From everything I have seen about work units distributed by E@H, they are extremely consistent in terms of the amount of computing required. I base this claim on the cpu time stats reported to me on all units validated over the past few weeks I have been here. For a given computer, the spread in CPU time looks to be less than 1%. E@H is very good in practice at making units that require a consistent amount of computation.
On the other hand, from everything I have seen about both claimed credits and granted credits, both vary far more & are far more unreliable & erratic than work unit computing required. I suppose this is due mainly to difficulties in calculating a reliable claimed credit with hyperthreading & dual CPUs & also difficulty amoung various operating systems & various hardware types.
So then, unless there is some plan or expectation to greatly vary the work unit computing requirement in the future, why fuss around with claimed credits at all? Why not say every work unit is worth, say, 80 credits if validated. That would not be perfect, but it would be subject to far less imperfection than the present claimed credit system, & would be simpler & easier to understand as well.
Why not make granted credit depend on something very consistent rather than something very inconsistent?
At least until the claimed credit calculation can be made more reliable than the work unit sizing. And it has a long way to go.
Fire away.
ADDMP
A modest proposal for granted credits...
)
> ... waiting to be shot down.
Not trying to shoot you down, just showing an example of how it isn't completely true that all units take the same amount of time.
Goto: http://einsteinathome.org/host/67977/tasks see that most of the units take one 30,000 seconds or so, but one of them only took 8,219
1943518 518223 13 Mar 2005 8:30:33 UTC 14 Mar 2005 10:51:41 UTC Over Success Done 8,219.53 23.30 94.42
such things just should not be writ so please destroy this if you wish to live 'tis better in ignorance to dwell than to go screaming into the abyss worse than hell
> > ... waiting to be shot
)
> > ... waiting to be shot down.
>
> Not trying to shoot you down, just showing an example of how it isn't
> completely true that all units take the same amount of time.
>
> Goto: http://einsteinathome.org/host/67977/tasks see that most of
> the units take one 30,000 seconds or so, but one of them only took 8,219
>
> 1943518 518223 13 Mar 2005 8:30:33 UTC 14 Mar 2005 10:51:41 UTC Over
> Success Done 8,219.53 23.30 94.42
Thanks for the response. I did look up that result of yours. The stats do report a very short CPU time for you as you say.
But did you look at the other two results there?
An Athlon64 3200+ took 22,927 CPU sec? Is that a short work unit? Is your computer really over twice as fast an an Athlon64 3200+?
And you were granted 94 credits. Not short either.
Just looks to me like a bad stat report on your results for some reason, & not an easy or short work unit. What do you think?
ADDMP (returning fire) :)
While the work is fairly
)
While the work is fairly close now I doubt this will always be the case. There are 3 different sized observatorys and so far all work has been from one run. These 2 factors make me expect randomly sized work in the future although the changes may fairly rare.
BOINC WIKI
BOINCing since 2002/12/8
> Not trying to shoot you
)
> Not trying to shoot you down, just showing an example of how it isn't
> completely true that all units take the same amount of time.
I noticed on a few of my wu's that the time occasionally resets when it changes from one project to another. This has resulted in a few of my results showing an extremely low time (and requesting extremely low credit also).
I actually have one that reports it did the work in 0.00 seconds, claimed 0.00 credits and was granted 64.34 credits. It returned a real result and it did pass validation - it just happened to hit that odd time reset right at the very end.
Work Unit 442042
Darren
> While the work is fairly
)
> While the work is fairly close now I doubt this will always be the case. There
> are 3 different sized observatorys and so far all work has been from one run.
> These 2 factors make me expect randomly sized work in the future although the
> changes may fairly rare.
Thanks. I don't know how E@H presently sizes work units, other than they have done it very tightly in what I have seen so far.
Are we doing mostly just Fast Fourier Transforming that depends mostly on the number of data points in the unit that is chosen as one package & sent out?
I agree that if the packaged units eventually become even more erratic (in computing required) than the present claimed credit calculation, then a flat credit scheme will not work well or fairly.
ADDMP